Table of Contents

AI 101 - How to build an AI FAQ agent

Raquel Updated by Raquel

This guide walks you through setting up a fully functional AI FAQ agent in Turn.io — from connecting an AI provider to handling escalations to a human. Follow along in your own Turn account, and you'll end up with a working agent you can customise to your organisation's needs.

Overview

Setting up an AI agent in Turn.io involves three stages:

  1. Integrating with an AI provider
  2. Creating an Assistant
  3. Setting up your Journey and Agent

The first two stages are one-time setup steps. Once they're done, you can reuse them across as many agents as you like.

Stage 1: Integrate with an AI provider

Turn.io uses a bring-your-own-key model. This means you create an account directly with the AI provider of your choice (for example, OpenAI), generate an API key there, and paste it into Turn.

  1. Go to Settings → AI
  2. Click AI models in the top right
  3. Find your preferred provider and paste in your API key

From this point on, that provider is available everywhere in the product.

Stage 2: Create an Assistant

An Assistant is a shared knowledge base for your agents. Think of it as the information you want every agent to know at all times — your tone of voice, organisation details, opening hours, locations, and so on.

You only need to create this once. Any agent you build later can reference it.

To create an Assistant:

  1. Go to Settings → AI
  2. Click Create new assistant
  3. Give it a name (for example, Turn Care)
  4. Select your AI provider and model — Turn pre-selects the provider you connected in Stage 1
  5. Fill in the fields:
    • Name and context — describe your organisation
    • Language — select the primary language your users communicate in
    • Tone of voice — for example: short, warm, professional, and clear
    • Capabilities — what users can do via this chat service
    • Objective — the main goal of your service
    • Knowledge base — any information your agents should reference when generating responses (clinic locations, FAQs, policies, etc.)
Keep your Assistant lean. Everything you add here is sent to the AI provider on every single message generation, which increases token usage and cost. Set your base tone and organisation context here, and add any agent-specific knowledge directly on the Agent (see Stage 3).

Stage 3: Set up your Journey and AI Agent

This is where you define how your agent actually behaves in conversation.

Add an AI Agent block to your Journey:

  1. Go to Journeys and open an existing Journey or create a new one
  2. Click + to add a new block and select AI Agent
  3. In the Agent settings, select the Assistant you created in Stage 2

The AI Agent block works like a back-and-forth conversation: the user sends a message, the agent replies, the user sends another message, the agent replies again — all within this single block.

Test your agent

Click Preview to open the simulator. Even with just the Assistant connected and no extra instructions, your agent already knows everything you defined — try asking it a question to see it in action.

Customise agent behaviour with prompts

The Agent prompt is where you define behaviour specific to this agent. While the Assistant holds shared knowledge, the Agent prompt defines what this particular agent should do differently.

Use the Agent prompt to:

  • Give the agent a specific greeting (for example: "Always greet the user with: Howzit! 👋")
  • Add knowledge that only this agent needs (keeping your Assistant lean)
  • Define constraints on what it should or should not discuss
Tip for prompt tuning: If the agent isn't behaving as expected, take a screenshot of the conversation, paste it into ChatGPT with your current prompt, and ask it to suggest improvements. Paste the revised prompt back into Turn and test again. Iterate until it behaves as intended.

Add an escalation action

Actions allow your agent to hand off the conversation to another part of your Journey — including a human agent. Escalation is one of the most common use cases.

To set up escalation
  1. In your AI Agent block, go to the Actions tab
  2. Click Add action and give it a name (for example: Escalate to human)
  3. Write a prompt that tells the agent when to trigger this action — for example:

    The user wants to speak with a person or has a medical emergency.

  4. Save the action
  5. After the action block, add a Text message to let the user know what's happening (for example: "We're routing you to our team. Hang tight.")
  6. Add a Route to operator block so the conversation surfaces in the Turn inbox for a human to pick up
Test escalation

In Preview, send a message like "I want to speak with a person." The agent should detect the intent, trigger the action, send your routing message, and surface the conversation in the inbox.

You're done — now make it yours

At this point you have a working AI FAQ agent. The next step is to replace the placeholder content with your real organisation's information:

  • Update the Assistant with your actual knowledge base — your real FAQs, locations, hours, and policies
  • Update the Agent prompt with any behaviour specific to your use case
  • Add more actions for other escalation paths or handoffs you need
What's next

This guide covers the essentials of getting an FAQ agent up and running.

In a follow-up guide, we'll cover more advanced capabilities — including how to use Skills to give your agent the ability to take actions on your behalf, such as booking appointments in a backend system.

Was this article helpful?

How to Get Your WhatsApp Template Reclassified from Marketing to Utility

Contact